Factors influencing severity of heavy-duty truck transport accidents based on LightGBM-SHAP

ObjectiveThe frequency of traffic accidents involving heavy-duty trucks has been on a continuous rise. Crashes in heavy-duty truck transportation often result in more severe consequences. Identifying the factors influencing severity of such accidents is therefore essential for the prevention of heavy-duty truck transport accidents.Method2 616 heavy-duty truck transport accident records were collected from the crash report sampling system database of United States from 2016 to 2020. The correlation analysis and recursive feature elimination algorithm were employed for feature selection. Fifteen potential factors influencing severity of heavy-duty truck transport accidents from seven dimensions (i.e., driver, vehicle, road, environment, time, space, and accident form) were extracted. Additionally, the class imbalance in training set was addressed by using adaptive synthetic sampling algorithm based on K-nearest neighbors. Four accident severity prediction models, i.e., LightGBM, XGBoost, random forest and SVM, were constructed based on data preprocessing. SHAP method was introduced to analyze the influencing mechanism of significant factors on accident severity.ResultThe proposed LightGBM model exhibits optimal overall performance. In terms of predictive accuracy, LightGBM model demonstrates superior performance with accuracy, F1 score, and AUC values of 0.872 1, 0.872 4, and 0.966 9 respectively. Regarding training speed, LightGBM model achieves the training speed of 7.65 s, which is more than 2.5 times faster than that with XGBoost, and notably faster than those with SVM and random forest, with speed advantages of 7 times and 16 times respectively.ConclusionThe SHAP-based model interpretation indicates that collision manners, unsafe driving behaviors, time of day, month, days of week, and roadway attributes are critical factors influencing accident severity. Among them, driving with the influence of alcohol or drugs, disregarding traffic signs or signals, severe fatigue driving, speeding, and distracted driving violation, as well as head-on and angle collisions, significantly contribute to the occurrence of severe injuries and fatal accidents. Moreover, the probability of fatal accidents involving heavy-duty trucks is higher during the time period of 0:00-4:00. These findings provide a theoretical foundation for accident prevention and the safety management of heavy-duty truck transport.